Previous meta-analyses identified an inverse association of coffee consumption with the risk of type 2 diabetes. However, an updated meta-analysis is needed because new studies comparing the trends of association for caffeinated and decaffeinated coffee have since been published.
PubMed and Embase were searched for cohort or nested case-control studies that assessed the relationship of coffee consumption and risk of type 2 diabetes from 1966 to February 2013. A restricted cubic spline random-effects model was used.
Twenty-eight prospective studies were included in the analysis, with 1,109,272 study participants and 45,335 cases of type 2 diabetes. The follow-up duration ranged from 10 months to 20 years. Compared with no or rare coffee consumption, the relative risk (RR; 95% CI) for diabetes was 0.92 (0.90–0.94), 0.85 (0.82–0.88), 0.79 (0.75–0.83), 0.75 (0.71–0.80), 0.71 (0.65–0.76), and 0.67 (0.61–0.74) for 1–6 cups/day, respectively. The RR of diabetes for a 1 cup/day increase was 0.91 (0.89–0.94) for caffeinated coffee consumption and 0.94 (0.91–0.98) for decaffeinated coffee consumption (P for difference = 0.17).
Coffee consumption was inversely associated with the risk of type 2 diabetes in a dose-response manner. Both caffeinated and decaffeinated coffee was associated with reduced diabetes risk.
Type 2 diabetes is a chronic disease with high rates of morbidity and mortality. The worldwide prevalence of type 2 diabetes is increasing, and the global number of people with diabetes is estimated to reach 366 million by the year 2030 (1). The risk of blindness, renal disease, and amputation among those with type 2 diabetes is 20 to 40 times higher than that of people without diabetes. Furthermore, those with type 2 diabetes have a two to five times higher risk of myocardial infarction and two to three times higher risk of stroke (2). Given its significant burden, identifying modifiable lifestyle factors is imperative for the prevention of diabetes.
Coffee is one of the most widely consumed beverages around the world; thus, investigating its association with various diseases has important public health implications. An inverse association between coffee consumption and risk of type 2 diabetes was first reported in a Dutch population (3). In subsequent years, this finding has been confirmed in most (4–10) but not all (11–13) studies. Two previous meta-analyses of coffee consumption and risk of type 2 diabetes have been published. van Dam and Hu (14) included nine cohort studies and reported a lower risk of type 2 diabetes for high coffee consumption compared with no coffee consumption with little between-study heterogeneity. Huxley et al. (15) included 18 prospective studies and found an inverse monotonous association between the number of cups of coffee consumed and diabetes risk, but they double-counted data from the same cohort (16,17). Since the publication of these meta-analyses, 10 additional prospective studies on the association between coffee consumption and diabetes have been published (6,8,10,12,18–23).
A key issue that remains to be resolved is whether consumption of caffeinated and decaffeinated coffee is similarly associated with the risk of type 2 diabetes. Such results would provide insight into the role of caffeine in the putative relationship between coffee consumption and diabetes risk. The meta-analysis by Huxley et al. (15) included only six studies on decaffeinated coffee, and estimates were less precise than for caffeinated coffee because of lower consumption levels. Eight subsequent prospective studies evaluated the association with decaffeinated coffee, approximately doubling the amount of data on decaffeinated coffee and diabetes risk (8,10,12,18–21,24). We therefore performed an updated systematic review and a dose-response meta-analysis of all available data on the association of both caffeinated and decaffeinated coffee consumption with the risk of type 2 diabetes.
RESEARCH DESIGN AND METHODS
Search Strategy and Selection Criteria
We searched the PubMed and Embase databases for prospective cohort studies or nested case-control studies that evaluated the association between coffee consumption and risk of type 2 diabetes between January 1966 and February 2013. The computer-based searches included the key words coffee and diabetes. No Medical Subject Headings terms were used because of the clear definitions of coffee and diabetes. Reference lists of retrieved articles were manually scanned for all relevant additional studies and review articles. We restricted the search to studies on humans and written in English.
Studies included in this meta-analysis met the following criteria: 1) the study design was prospective cohort or nested case-control; 2) the articles were published in English; 3) the exposure was categorized coffee consumption, including total coffee, caffeinated coffee, or decaffeinated coffee; and 4) the outcome was risk of type 2 diabetes. Studies were excluded if they were cross-sectional in design and if information on dose-response modeling was inadequate, including the number of participants and cases, relative risk (RR) and SE for the estimate, and dose of coffee consumption in each exposure category.
Data Extraction and Quality Assessment
One author (M.D.) assessed study eligibility and extracted the data, and another (M.C.) independently double-checked the available data. The following data were extracted from each study: first author’s name, year of publication, geographical location, follow-up time, sex, age, number of type 2 diabetes events, number of participants, categories of coffee consumption, type of assessment of type 2 diabetes, covariates adjusted for in the multivariable analysis, and RRs for all categories of coffee consumption. If studies based on the same cohort were published several times, only the most recent article was selected. If a study only reported caffeinated coffee consumption instead of total coffee consumption, caffeinated coffee consumption was also included in the total coffee consumption analysis. The Newcastle-Ottawa quality assessment scale (NOS) was used to evaluate the study quality of the included articles. The scale ranges from 0 to 9 points, with higher points indicating higher study quality. Specifically, loss to follow-up was assessed as an item in the NOS for individual studies. In the present meta-analysis, we assumed that loss to follow-up was random.
To perform a dose-response meta-analysis, we assigned the median or mean coffee consumption in each category of consumption to the corresponding RR for each study. If the mean or median consumption per category was not reported, the midpoint of the upper and lower boundaries in each category was assigned as the mean consumption. If the upper boundary for the highest category was not provided, the assigned mean value was 25% higher than the lower boundary of that category. If the lower boundary for the lowest category was not provided, the assigned mean value was one-half of the upper boundary of that category.
Data Synthesis and Analysis
To analyze the trend of coffee consumption and risk of type 2 diabetes, we used both semiparametric and parametric methods. For the semiparametric method, four coffee consumption groups were generated, namely lowest, third highest, second highest, and highest. For each included study, the lowest and the highest coffee consumption categories corresponded to the lowest and highest groups, respectively. For studies with four exposure categories, the second and third categories corresponded to the second and third highest groups, respectively. For studies with three exposure categories, the middle category corresponded to either the second or the third highest group, whichever median coffee consumption amount was most similar. If the study had more than four exposure categories, two consumption groups, other than the lowest and highest, were chosen on the basis of their similarity of the amount of coffee consumption in that category to the second and third highest groups. For each group, a random-effects model was used to pool the RR of type 2 diabetes (25). Statistical heterogeneity in each group was assessed by Cochrane Q test, with P < 0.1 indicating significant between-study heterogeneity.
For the parametric method, a random-effects dose-response meta-analysis was performed (26). The number of cases and participants in each coffee consumption category was extracted to estimate the covariance of the RR in each study. Together with the observed adjusted variance of the RR, we estimated the variance/covariance matrix of the data. The weight of each study was calculated as the inverse of the variance/covariance matrix. We used generalized least squares models with maximum likelihood method to estimate the coefficients for each study, and we fit a random-effects generalized linear model. Additionally, we tested for potential nonlinearity in the association between coffee consumption and type 2 diabetes by a random-effects restricted cubic spline model with three knots. Stata commands GLST for model fitting and LINCOME for estimating effect were used for the fitted model.
Stratified analyses were performed for study location, sex, diabetes assessment method, and type of coffee (i.e., caffeinated, decaffeinated), and the subgroup difference was tested in the association of coffee consumption with risk of type 2 diabetes. We assessed the potential for publication bias through a formal Egger regression symmetry test. All analyses were conducted with Stata 10 (StataCorp, College Station, TX) statistical software.
Characteristics of Studies
The initial search identified 800 potentially relevant citations. After screening titles and abstracts, we identified 39 studies for further evaluation (Fig. 1). We excluded six studies with a cross-sectional study design, three that were abstracts of conference posters, and two that were repeated in the same cohort. The results of the remaining 28 studies, comprising 1,109,272 study participants and 45,335 cases of type 2 diabetes, were included in the meta-analysis (3–13,17–24,27–35). Characteristics of all 28 studies were shown in Table 1. The duration of follow-up for incident type 2 diabetes ranged from 10 months to 20 years, with a median follow-up of 11 years. Thirteen studies were conducted in the U.S., 11 in Europe, and 4 in Asia. In nine studies, type 2 diabetes was self-reported; the outcome of six studies was assessed by means of a glucose tolerance test, and the outcome of the other studies was confirmed by either medical records or national registries. One study was a nested case-control design, and the remaining 27 were prospective cohort studies. Ten studies assessed both caffeinated and decaffeinated coffee, and only one each assessed caffeinated coffee and decaffeinated coffee. Fifteen studies did not distinguish caffeinated coffee and decaffeinated coffee, and seven assessed caffeine consumption. The mean NOS score was 7 (of a possible 9 points), suggesting a high quality of the studies included in the meta-analysis. The points were mainly lost in exposure assessment and adequacy of follow-up of cohorts: Six studies assessed coffee consumption by structured interview, and five addressed the percentage of loss to follow-up.
Total Coffee Consumption and Risk of Type 2 Diabetes
Figure 2 shows the results of different levels of total coffee consumption compared with the lowest category. Compared with the lowest category (median consumption 0 cups/day), the pooled RR for incident type 2 diabetes for individuals in the highest category of consumption (5 cups/day) was 0.70 (95% CI 0.65–0.75, I2 = 50%, P for heterogeneity = 0.001). The corresponding RRs were 0.80 (0.75–0.85, I2 = 71%, P < 0.001) for the second highest category (3.5 cups/day) and 0.91 (0.88–0.94, I2 = 19%, P = 0.17) for the third highest category (1 cup/day) of coffee consumption. Thus, there was evidence for substantial between-study heterogeneity in results for the highest two categories of coffee consumption.
Dose-Response Analyses of Total Coffee Consumption and Risk of Diabetes
We performed a dose-response meta-analysis in 27 studies to test for a linear trend between total coffee consumption and risk of type 2 diabetes and to estimate RRs for specific amounts of coffee consumption. One study included coffee consumption of >12 cups/day for the highest category (23); we treated this observation as an outlier and excluded it from the dose-response analysis. There was a strong inverse association between coffee consumption and risk of type 2 diabetes (Fig. 3). A cubic spline model accounted for more variance in the outcome than did a linear model (likelihood ratio test P < 0.001), suggesting that the association was not fully linear. Compared with participants with no coffee consumption, the RR estimated directly from the cubic spline model for 1–6 cups/day was 0.92 (95% CI 0.90–0.94), 0.85 (0.82–0.88), 0.79 (0.75–0.83), 0.75 (0.71–0.80), 0.71 (0.65–0.76), and 0.67 (0.61–0.74), respectively.
To test whether the inverse association between coffee consumption and risk of type 2 diabetes was different for unadjusted and adjusted RRs, we performed a dose-response meta-analysis of the unadjusted data in 27 studies (Supplementary Fig. 1). The spline curve of the unadjusted data was similar to that of the multivariable-adjusted data, indicating that adjustment for potential confounders minimally affected effect estimates for the association between coffee consumption and a lower risk of diabetes.
Comparison of Associations for Caffeinated and Decaffeinated Coffee
Both caffeinated and decaffeinated coffee consumption was inversely associated with risk of type 2 diabetes (Supplementary Fig. 2). Compared with participants with the lowest level of caffeinated coffee consumption, the RR for incident type 2 diabetes was 0.74 (95% CI 0.67–0.81, I2 = 56%, P for heterogeneity = 0.005) for the highest category (median consumption 5 cups/day), 0.82 (0.75–0.91, I2 = 74%, P < 0.001) for the second highest category (3.5 cups/day), and 0.92 (0.89–0.96, I2 = 6.6%, P = 0.38) for the third highest category (1 cup/day). Compared with those with the lowest level of decaffeinated coffee consumption, the combined RR for incident type 2 diabetes was 0.80 (0.70–0.91, I2 = 62%, P = 0.001) for the highest category (4 cups/day), 0.95 (0.88–1.02, I2 = 53%, P = 0.02) for the second highest category (2 cups/day), and 0.98 (0.92–1.05, I2 = 62%, P = 0.003) for the third highest category (0.5 cups/day). Heterogeneity was shown for five of the six subgroups. Seven of 12 studies adjusted for caffeinated and decaffeinated coffee consumption simultaneously. The association was slightly stronger for caffeinated coffee consumption than for decaffeinated coffee consumption (P = 0.03 for the second highest group, P = 0.07 for the highest group) (Supplementary Table 1).
We conducted a linear dose-response analysis for caffeinated and decaffeinated coffee consumption separately (11 studies). For a 1 cup/day increase in caffeinated coffee consumption, the RR for type 2 diabetes was 0.91 (95% CI 0.89–0.94), and for a 1 cup/day increase in decaffeinated coffee consumption, the RR was 0.94 (0.91–0.98, P for difference = 0.17). We performed a dose-response analysis between caffeine consumption and type 2 diabetes risk based on seven of the included studies (4,9,12,17,21,22,35) and found that for every 140 mg/day (∼1 cup/day coffee) higher caffeine consumption, the RR for type 2 diabetes was 0.92 (0.90–0.94). Of note, none of the included studies controlled for coffee intake when modeling caffeine intake and diabetes.
In stratified analyses, the inverse associations between coffee consumption and risk of diabetes were similar by geographical region (U.S., Europe, and Asia), sex, and diabetes assessment method (P for interaction > 0.05 for all groups) (Supplementary Table 1).
For the dose-response analysis of total coffee consumption with diabetes, after we included the observation with extremely high coffee consumption (23), the results did not change. Significant heterogeneity was found for the highest and second highest total coffee consumption, which might be attributable to the heterogeneous amount of coffee consumption in these categories. The dose-response analysis accounted for the heterogeneous consumption amount of individual studies and showed an appropriate goodness of fit (P = 0.14).
Heterogeneity was also found in separate analyses of caffeinated and decaffeinated coffee consumption. In addition to considering different consumption amounts, we stratified the analyses by whether the studies included caffeinated and decaffeinated coffee consumption simultaneously in the same model. In these analyses, the heterogeneity at each level of caffeinated and decaffeinated coffee consumption decreased significantly, with even more inverse results for both caffeinated and decaffeinated coffee consumption (Supplementary Fig. 3).
The Egger test provided no evidence of publication bias at any levels of total coffee, caffeinated coffee, and decaffeinated coffee consumption (Supplementary Table 2).
The findings from this systematic review and meta-analysis, based on 1,109,272 study participants and 45,335 cases of type 2 diabetes, demonstrate a robust inverse association between coffee consumption and risk of diabetes. Compared with no coffee consumption, consumption of 6 cups/day of coffee was associated with a 33% lower risk of type 2 diabetes. Caffeinated coffee and decaffeinated coffee consumption were both associated with a lower risk of type 2 diabetes. The association between coffee consumption and diabetes risk was consistent for men and women and for European, U.S., and Asian populations.
There has been some debate about the role of caffeine in the development of insulin resistance and diabetes. Short-term studies in humans have shown that caffeine intake results in an acute reduction of insulin sensitivity, which may be due to adenosine receptor antagonism and increased epinephrine release caused by caffeine (36). Other studies have found that when controlling for total coffee intake, caffeine intake was no longer associated with diabetes risk (5,19). Nonetheless, these results need to be interpreted with caution because the high correlation between coffee and caffeine consumption makes it more difficult to separate their effects. The present meta-analysis shows that both caffeinated and decaffeinated coffee consumption is associated with a lower risk of diabetes. The results were more inverse when only including the studies adjusting for caffeinated and decaffeinated coffee simultaneously in the model. Because caffeinated coffee consumption is inversely related to decaffeinated coffee consumption, caffeinated coffee is a positive confounder for the association of decaffeinated coffee with diabetes and vice versa. Thus, the association is strengthened when both variables are adjusted simultaneously. Moreover, because the consumption of caffeinated coffee is much higher than decaffeinated coffee, the results appear to be more robust for caffeinated coffee. However, the difference for continuous analysis between caffeinated and decaffeinated coffee (1 cup/day increment) was not statistically significant. These results suggest that components of coffee other than caffeine are responsible for this putative beneficial effect. Although the present results show that higher caffeine consumption is inversely associated with the risk of diabetes, this association is likely to be confounded by other components of coffee because of the collinearity of caffeine and those components.
Plausible biological mechanisms that may contribute to the inverse association between coffee consumption and diabetes risk have been demonstrated in animal models and in vitro studies implicating several coffee components in reducing insulin resistance and improving glucose metabolism. Chlorogenic acid, a phenolic compound, is a major component of coffee and has been shown to reduce blood glucose concentrations in animal experiments (37). Chlorogenic acid may reduce glucose absorption in the intestines by competitively inhibiting glucose-6-phosphate translocase and reducing sodium-dependent glucose transport in the brush border membrane vesicles (38), by reducing oxidative stress as a result of its antioxidant properties, and by reducing liver glucose output (39). Other components of coffee may also improve glucose metabolism, including lignans, quinides, and trigonelline (40–42). Coffee is also a good source of magnesium, which has been associated with a lower risk of type 2 diabetes (43). However, adjustment for magnesium intake did not substantially attenuate the inverse association (4,5), suggesting that other bioactive compounds in coffee may also be responsible. In addition, because some of the compounds, such as lignans and magnesium, also exist in other foods, it is difficult to attribute the benefits of coffee on diabetes to these specific compounds only.
This meta-analysis has several significant strengths. The mean NOS score of 7 of 9 ensured the relatively high quality of the included studies. All 28 studies used a prospective design; thus, the differential misclassification of coffee consumption attributable to recall bias was minimized. Given the observational nature of the studies, residual confounding might be an explanation for the findings. However, higher coffee consumption was generally associated with a less healthy lifestyle, including a high prevalence of cigarette smoking, less physical activity, and a less healthy diet. Thus, the true association between coffee and diabetes risk might be stronger than observed. Indeed, the similarity found in the present adjusted and unadjusted results indicates that the association is unlikely to be substantially confounded by known diabetes risk factors. Nonetheless, it is difficult to establish the causality between coffee consumption and diabetes solely based on observational evidence.
Because most of the studies coffee was assessed through self-reported dietary questionnaires, misclassification of coffee intake is inevitable. However, the misclassification is likely to be nondifferential with respect to the outcome because of the prospective design of the studies, which might have attenuated the observed association. Only five studies addressed the loss to follow-up percentage. However, the loss to follow-up is unlikely to be associated with coffee consumption and may have resulted in selection bias. Coffee brewing methods were not assessed in the studies. However, most coffee is likely to be filtered coffee, and the results from studies conducted in various populations, including U.S., European, and Asian, were similar, indicating consistency of the results despite potentially different preparation and processing methods. In addition, none of the studies assessed the amount of sugar and dairy added to coffee. However, the amount of sugar and dairy added to coffee is likely to be small compared with other food sources. Finally, because of the observational nature of the studies, a causal relationship cannot be established with these data alone.
In conclusion, this meta-analysis provides strong evidence that higher consumption of coffee is associated with a significantly lower risk of diabetes. Both caffeinated coffee and decaffeinated coffee are associated with reduced diabetes risk. Longer-term randomized controlled trials are needed to establish causality and to elucidate the underlying mechanisms.
Funding. This research was supported by National Institutes of Health grant DK-58845.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. M.D. contributed to the data research, extraction, and analyses and wrote the manuscript. M.D., S.N.B., M.C., R.M.v.D., and F.B.H. contributed substantially to the study conception and design, data analysis and interpretation, and drafting and critical revision of the manuscript for important intellectual content. S.N.B. contributed to the manuscript organization and reviewed and edited the manuscript. M.C. double-checked the extracted data and reviewed the manuscript. R.M.v.D. and F.B.H. contributed to the introduction and discussion and reviewed and edited the manuscript. F.B.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.